Anand: PII vs. Q9550 vs. i7 crossfire, Phenom II = smoother

[Mav451]

If we indeed are going to get scientific about it, you need to understand that FPS inherently IMPLIES an average (frames every second). We need something more precise, where we can get frame-rate down to something like every 100ms. Then, at this level of precision, you can use the derivative of those peaks/valleys and come to an actual numeric value that would accurately convey the "smoothness", if you will, of said system.

[Rammsteiner]

just look back before the release of P2, even there you had the "smoother" argument.

In my eyes the whole "smoother" debate is a non issue, if people have concerns about "smoothness" just tell them to enable vsync/frame rate limiter and everythings fine.

To be honest, I doubt that's actually true but well

Maybe it's indeed in the fps, but it might be caused by interconnect with CPU and GPU or whatever. I dont know.

[Dzigit]

I think, that problem of smoothness is not a problem of less framerate variance, because you cant see any difference if you have more than about 70-80 frames per second. As Mav451 noticed FPS is average so there is a possibility that in first half of the second we will have 50 fps, and in another half of the second we will have 100 fps, so it gives us average 75 fps, with which we shouldnt see any lags, but in first half of second we can see it.

I think in benchmarks we should look at max and min time between frames. If our max time between frames is less than about 12 ms then we shouldnt see any lags.

So if intel machines gives us higher max time between frames (MTBF ? ) then it means that AMD is smoother.


P.S. My first post

[keithlm]

So if intel machines gives us higher max time between frames (MTBF ? ) then it means that AMD is smoother.


P.S. My first post

Welcome to the forum.

What you call the MTBF is what I called the delta time in my post #273.

If you graph this number versus the frame so that you can see the min and max then I think that it might actually display something that is measurable and repeatable.

This would not just be personal opinion and could not just be laughed off as something that doesn't exist. (Although some posters will attempt that even if this method does produce tangible results.)

Actually this is just good science. If we have a number of people experiencing something that we don't currently measure using the currently accepted methods then we need to find methods or a new model which explains and confirms or denies the experiential data. But going by past experience now that I've actually said that we want to find a method to measure this attribute we can expect several posts claiming it is not measurable and is subjective only.

[iandh]

Welcome to the forum.

What you call the MTBF is what I called the delta time in my post #273.

If you graph this number versus the frame so that you can see the min and max then I think that it might actually display something that is measurable and repeatable.

This would not just be personal opinion and could not just be laughed off as something that doesn't exist. (Although some posters will attempt that even if this method does produce tangible results.)

I think the average of the top ten percent of maximum time between frames would be a great tangible measure of "smoothness", and quite irrefutable IMO.

[CntdwnToExtn]

Well, my worthless cents added...

If i7 isn't affordable at the $1500 price tag, why not just flip a coin between the PII/Q9550? Both are at $1000 budgets.
This is assuming you have a whole new build.

I mean, looking at the FPS AMD is either right there (limiters) or less than 10fps behind (minus the L4D 1920) - at resolutions our eyes can't even tell the difference except when down on paper.
If you can tell the difference between 119fps and 110fps time after time after time again...I would love to see that. Honestly would. It would be pretty cool to see.
But to 'nit pick' about who's smoother or has more fps by 2...worthless chat.
It's not like it's an extra 2mpg where over years it will save you quite a few tanks of gas for a commuter. It's a frame rate that goes down onto a piece of paper...to sit there until either the ink or paper deteriorates.

Anyhoo,
If you have money in or work for either Intel or AMD, I find absolutely no problem defending the company.
But come on people, they are not monitoring these forums looking for people.
Calm down, drink a beer, smoke something, use your free hand...something to release the built up tension.
It's not like AMD or Intel killed someone you know.

[keithlm]

I think the average of the top ten percent of maximum time between frames would be a great tangible measure of "smoothness", and quite irrefutable IMO.

You are probably correct. But I think the change in the change would be more important. (OMG we just discovered benchmark calculus.)

But first we have to determine if this method creates something that can be seen as being different between the architectures. If it can't then we move on and look for something else that might.

[roofsniper]

If we indeed are going to get scientific about it, you need to understand that FPS inherently IMPLIES an average (frames every second). We need something more precise, where we can get frame-rate down to something like every 100ms. Then, at this level of precision, you can use the derivative of those peaks/valleys and come to an actual numeric value that would accurately convey the "smoothness", if you will, of said system.

fraps has frame times. just like the excel sheet i showed.

@ keithlm on that pic you posted with the two graphs you can see that one stays about the same and one varies up and down. the one that varies has about the same average, has a higher max and has the same min. but thats not what you want to see. that is the idea of what kind of chart i would like to see. a chart showing the times it took the frames to be produced over time. with the time taking for each frame to be a data point. like you said in a perfect situation it should be a straight line.

[Sparky]

fraps has frame times. just like the excel sheet i showed.

@ keithlm on that pic you posted with the two graphs you can see that one stays about the same and one varies up and down. the one that varies has about the same average, has a higher max and has the same min. but thats not what you want to see. that is the idea of what kind of chart i would like to see. a chart showing the times it took the frames to be produced over time. with the time taking for each frame to be a data point. like you said in a perfect situation it should be a straight line.

Get me the data and I can chart that for you.

[roofsniper]

Get me the data and I can chart that for you.

well first we need to have two systems which gave about the same average fps. one a core 2 and one a phenom I or II.

[Mav451]

fraps has frame times. just like the excel sheet i showed.

Haha oops. Well, then I guess all we're waiting for is someone to get SparkyJJO numbers to crunch.

[roofsniper]

Get me the data and I can chart that for you.

hey im curious about doing it for myself but i have rarely used excel. so to do this you would need to get one cell to subtract the value from the cell above it to get the time it took to get that frame. what would i need to type to get it to do this over and over? cause its a pita to tell it to get b4-b3 then b5-b4, etc.

[Sparky]

hey im curious about doing it for myself but i have rarely used excel. so to do this you would need to get one cell to subtract the value from the cell above it to get the time it took to get that frame. what would i need to type to get it to do this over and over? cause its a pita to tell it to get b4-b3 then b5-b4, etc.

You can do a click-drag from the bottom right corner of the cell that has the forumula. So you would drag it across or down and it would automatically fill in the cells and increment the references.

[keithlm]

hey im curious about doing it for myself but i have rarely used excel. so to do this you would need to get one cell to subtract the value from the cell above it to get the time it took to get that frame. what would i need to type to get it to do this over and over? cause its a pita to tell it to get b4-b3 then b5-b4, etc.

Sorry... I've been busy doing actual work stuff today. (OH THE TRAVESTY.)

Anyway... using the data you showed us we get this:
(Of course this only 48 frames... we should use thousands... but you get the idea. And it still looks like this idea might be useful.)

[roofsniper]

You can do a click-drag from the bottom right corner of the cell that has the forumula. So you would drag it across or down and it would automatically fill in the cells and increment the references.

nice thanks. anyway so we need to find two similar systems now.

[Hornet331]

If we indeed are going to get scientific about it, you need to understand that FPS inherently IMPLIES an average (frames every second). We need something more precise, where we can get frame-rate down to something like every 100ms. Then, at this level of precision, you can use the derivative of those peaks/valleys and come to an actual numeric value that would accurately convey the "smoothness", if you will, of said system.

i very much support this, are there any progamers there that can write an app that can log frames over a variable time (frames per 100ms or even better frames per ms).

But loging in this short intervalls will generate much data, for 100ms you have 10 datasets per second and if you log each ms you will get 1000 datasets per second, should be made sure that the logging is on a extra drive or ramdisk so it dont influences the game itself.

Maybe it's indeed in the fps, but it might be caused by interconnect with CPU and GPU or whatever. I dont know.

Watch it your making the same assumption as gosh did, and jj alrady showed that theres next to no difference in games, regardless if its a HT/IMC bus or FSB that is used, even a FSB of 200 mhz is sufficient to deliver the same fps as a stock C2 or a P1/P2 with HT.

<sarcasm mode ON>
We're all just shocked that you, of all people, find this to be a non-issue.
<sarcasm mode OFF>

This is most especially true since we almost have a constructive conversation going on. That can NOT be allowed. You'll need to run to the other forum subsection and gather some of your cohorts.

Actually I'm sure I speak for many people when I say your comment is worthy of a very large yawn.

And changing the vsync won't actually change anything. It might hide the issue from any method that could possibly be used to determine the difference. (But then hiding the issue is probably acceptable to some people.)

and im not surprised that statement comes from you.

I am constructive, i give you the data i know off, concerning this data. Im intrigued by this yet your only statement is a dull flame towards me. But im not surprised at all, since you already labeled me as biggest troll in this forum.

Tell my why its a issue if you can "fix" that issue with enabling vsync/a frame limiter.
I am a gamer, i tweak my system/games to be as fast/pretty looking and yes even as smooth as possible. You lable it as cheat so that i can troll the forum, i lable it as way to get the best out of the games i play.

@ topic
some thought i have come up:

1)
If the perceived "smoothness" originates form high differences between min. and max. fps then the Cpu with the most raw performance is prone to this phenomenon. (especial in none gpu limted scenarios, cause most of the minimus are caused by graphics, only sometimes by the cpu)

2)
If the perceived "smoothness" originates from a non steady fps rate, it hard to tell right now, cause there aren't precise enough tools to measure it (we dont see fluctuations with the tools we have now), e.g. right now available tools only can measure the fps once per second, and if the frames would fluctuate over a larg ammount are in between this second, this might be the source of this phenomenon.

3)
If the perceived "smoothness" originates from the "eye of the beholder", as in the much debated topic "you can't see more then xx fps", the whole smoothness phenomenon may is preciceived different by each person. E.g. the data will be the same, and its the user/gamer that reads something into this.

That is what i have come up so far, regarding to this topic over the time and variouse other topics concerning this phenomenon.

1) and 2) are provable by tools and 3) is not provable in any way.

I hope someone comes up with a tool as Mav451 suggested.

[KELL5]

hmmmm I seem to recall a pretty important person 'round here sigged

Phenom + Bull = Smoother

Can't for the life of me remember who though

[roofsniper]

i very much support this, are there any progamers there that can write an app that can log frames over a variable time (frames per 100ms or even better frames per ms).

i have said it at least 10 times in this thread and in many other threads. fraps logs the frame times that each frame comes out and thats as precise as it gets. all you do then is find the amount of time within each frame and make a chart. but first we need two systems that get the same exact average fps and then develop a testing method to do this.

[Hornet331]

but first we need two systems that get the same exact average fps and then develop a testing method to do this.

ask jumpingjack, he has identical rigs (of course cpu/mobo is different) and if you ask nicely he will supply you with data, i think he even has a P2 now, afair he even said he might do something like that between P1,P2, Ci7 and C2D.

[*GameOver*]

I can get you all the data you want from a 9950 system with a 280GTX.. But i dont have time at work to sit and collect this data off our head to head machines. Im far to busy sadly. Let me know what game you want to use and ill try and fraps the data out of it that your looking for. if someone els has a C2D or C2Q machine with a 280gtx and similar ram configuration that would be helpful.

[Mav451]

Meh, I don't think average FPS matters at all. After all, the most important thing is that the derivative between the peaks/valleys is (absolute value) as close to 0 as possible. Overall performance, it doesn't matter - all we need right now is AMD or Intel users to start gathering data to generate this stuff. I guess the more important question:

What game are we using? And we need, at the very least, to keep the GPUs/RAM timings/etc. as similar as possible. Probably be easier for someone with both AMD and Intel setups.

[Hornet331]

I can get you all the data you want from a 9950 system with a 280GTX.. But i dont have time at work to sit and collect this data off our head to head machines. Im far to busy sadly. Let me know what game you want to use and ill try and fraps the data out of it that your looking for. if someone els has a C2D or C2Q machine with a 280gtx and similar ram configuration that would be helpful.

Same basic hardware is imho not enough, you need same HDD, same addin cards (soundcard other gimmicks), same os, running same background services, same resolutions, same game settings, etc. (you get the idea )

[keithlm]

Here are two new charts.

The first one is A quick world of warcraft run.

The second is basically the same thing WITH 4xPrime95 running at the same time.

It is obvious that the first run is "smoother" than the second run. Although I wish I had stretched them both out the same. (Okay... I re-saved the second one.)

Note: REMEMBER that this is NOT FPS. It is time between frames. The FPS on the first run was between 200 and 300 fps. The FPS on the second run was around 50FPS.






(And yes Hornet you ARE a troll.)

[roofsniper]

Same basic hardware is imho not enough, you need same HDD, same addin cards (soundcard other gimmicks), same os, running same background services, same resolutions, same game settings, etc. (you get the idea )

of course. the results are most likely gonna have to come from a person with both systems.

[Hornet331]

(And yes Hornet you ARE a troll.)

lol ninja edit to save your face, cause saying its a none issue, if you enable vsnyc, is so much of trolling....

of course. the results are most likely gonna have to come from a person with both systems.

as said, try to ask JJ he has the means, and im sure when some people ask he might also has the will.